The CrimeRisk score for my area is 208. What does that actually mean?
CrimeRisk scores are indexed to the national level, which is given a score of 100. A value of 100 for a local area means that its expected risk is about equal to the national average. A score of 200 indicates that we expect double the risk of crime, and a score of 50 would indicate one half the national risk.
It is important to realize that these are not normally distributed (bell curve) but are highly skewed. There are many more areas with scores below 100 than with scores above, because most crimes tend to be highly geographically concentrated.
It is also vital that users understand that these are relative indexes, and as such contain error arising from problems with data collection and statistical model performance. Therefore, we strongly advise against having “cut-off” values when working with these scores since the reality of a neighborhood with a score of 199 is not noticeably different than one with a score of 200. Instead, we prefer that people work with ranges – a score under 100 means “not a concern”, a score over 500 means “strong concern”, and anything in between indicates that further review, or consideration of other factors, needs to be brought to bear on the problem at hand.
How are total crimes, personal crimes, and property crimes indexed?
There are seven base crime indexes, which can be classified in the following hierarchy –
- Total Crime
- Crimes against Persons
- Crimes against Property
- Motor Vehicle Theft
- Crimes against Persons
Each of the seven base crimes are modeled independently as the spatial patterns vary substantially. Interested readers should peruse the detailed documentation on methodology here [link to methodology].
In building the summary categories, we avoid trying to explicitly weight incidents by type – meaning that for building the property crime variable, we simply add the predicted values for each core crime and index those to the national average. In effect, the implicit weight given is that all criminal incidents, regardless of type, are of equal importance. This in fact follows the standards set by the FBI Uniform Crime Reports rather than attempting so set our own standards – which would involve making sweeping and judgement based claims that each murder is worth a certain number of robberies, assaults, and/or motor vehicle thefts.
How Detailed is CrimeRisk?
For most users, the block group (220,000 areas) represents an area which is sufficiently detailed but not unmanageable in terms of data storage and processing and is usually the preferred geography for delivery. In our view, the larger the area, the less useful the estimates become, since crime rates in adjacent neighborhoods can vary tremendously in many cities. For those interested in more detailed statistics, CrimeRisk is modeled first at the census block level (several million areas) and is available to users at this and the ZIP+4 level should clients require data for these micro-areas.
Is There Racial or Ethnic Bias in the Data?
It is important to understand that AGS does not use any variables related to race, Hispanic origin, ancestry, or language spoken at home in constructing the models of each crime type. The variables instead focus on basic demographics (age and sex), family structure, and socio-economic characteristics (income, education, employment). Typically, the variables which best explain spatial distributions of crime are related to two main factors – the percentage of the population male aged 18-25, and the percentage of families that are lone parent with children.
There are, however, correlations between the spatial patterning of crime and racial/ethnic patterns which reflect the reality of crime in our major cities. It is important to stress that these spatial correlations are not as a result of explicitly biasing our models by using such variables.
Can We Simplify CrimeRisk For Our Users?
Some users prefer a simpler approach with just a few categories, typically given labels like “very safe”, “safe”, and so on. The AGS data delivery platforms both allow users to reclassify a variable using the CLASS() function. So, if you want a crime index that ranges from 1 to 5, you can easily specify something like CLASS(CRMCYTOTAL,5,80,120,200,500) which would reclassify the total crime index as:
Index < 80 1
Index 80-119 2
Index 120-199 3
Index 200-499 4
Index >=500 5
It is also possible to re-cast the indexes as percentiles, deciles, or quartiles, where areas are given a score such as “Riskiest 10%” should that be desirable.
If you have other questions, please contact firstname.lastname@example.org.